15
Qualication of an ultrasonic ow meter as a transfer standard for measurements at Reynolds numbers up to 4 10 6 between NMIJ and PTB L. Cordova a,n , N. Furuichi b , T. Lederer a a Physikalisch-Technische Bundesanstalt, Germany b National Institute of Advanced Industrial Science and Technology, Japan article info Article history: Received 19 June 2014 Received in revised form 7 April 2015 Accepted 19 April 2015 Available online 24 April 2015 Keywords: Interlaboratory comparison Ultrasonic ow meter Reynolds number dependence Flow traceability Transducer cavity abstract The quality of any laboratory intercomparison depends to a large extent on the performance of the used ow meter. To nd a ow meter that is capable of reaching a reproducibility better than 0.05% requires bounding all involved inuence quantities down to the required level. The present paper describes the efforts performed while qualifying a time-of-ight ultrasonic ow meter as a transfer standard. It was determined that the most relevant inuence quantity besides the ow prole within the bulk ow is the effect caused by the transducer pockets in the meter body. By taking advantage of a specially designed window chamber, it was possible to determine the magnitude of the errors introduced by the transducer pockets and to dene, based on the ndings, a procedure to perform a bilateral comparison between the hot water calibration facilities of the Physikalisch-Technische Bundesanstalt and the National Institute of Advanced Industrial Science and Technology. The results of the bilateral comparison are presented. & 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). 1. Introduction and motivation Water is used as an energy transporting medium in every type of power plant involving turbines; also industrial and district heating depend on accurate measurements of ow rate. In most cases, the actual measurement uncertainty is in the order of 1%. Consequently, every improvement of the measurement uncertainties has direct consequences for the safety and efciency of the involved systems. Flow rate measurements in the eld are performed ideally by instruments that have been tested at National Metrology Institutes (NMI) or at a calibration laboratory that has been accredited and/ or is participating in prociency tests organized by the corre- sponding NMI as can be seen in Fig. 1. Any bias introduced by a calibration laboratory would have a direct impact on the price, on the quality or on the competitiveness offered by its clients. In order for measurements to be globally consistent, it is required that NMIs prove their mutual consistency periodically through international comparisons. The Mutual Recognition Arrangement of the International Committee for Weights and Measures (CIPM- MRA) has established mechanisms in order to allow the NMIs to prove their mutual consistency transparently and based on the same rules and principles. Actually there are more than 53 states and 152 institutes, designated by the signatory bodies, participat- ing in the CIPM-MRA. The traceability of a ow rate calibration facility is normally assessed on a quantity-based calibration, i.e. mass, volume, time, density and temperature standards are calibrated separately. Only in cases where there is a ow meter capable of delivering reproduc- ibilities much lower than the required calibration uncertainties it is possible to provide a direct ow-rate traceability. This is possible in low-ow hydrocarbon measurements as reported by Shimada. Highly reproducible measurement instruments are available as seen, for example, at the Calibration Intercomparison on Flow Meters for Kerosene carried out on 1995 [10] and the CIPM-MRA international key comparison of liquid hydrocarbon ow facilities CCM-FF-K2 [11]. Without direct ow-rate traceability, systematic errors in any system of the calibration rig might remain undetected. There are several relevant ow rate measurements in the eld performed without a calibration as depicted in Fig. 1. This situation is given mostly in cases where the measurement conditions cannot be reproduced in a laboratory. Under these circumstances the only alter- native is to apply ow measurement technology that has a predictable working principle that allows the use of similarity principles to infer the calibration result and uncertainty of measurements under conditions different from those present during calibration. The relevant ranges for energy transport through hot water vary mainly between 50 1C and 250 1C. Flow rates larger than 3500 m 3 /h have been reported and Reynolds numbers up to 30 10 6 . According Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/ owmeasinst Flow Measurement and Instrumentation http://dx.doi.org/10.1016/j.owmeasinst.2015.04.006 0955-5986/& 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). n Corresponding author. E-mail address: [email protected] (L. Cordova). Flow Measurement and Instrumentation 45 (2015) 2842

Journal MGP

Embed Size (px)

DESCRIPTION

Diario hidraulico

Citation preview

  • Qualication of an ultrasonic ow meter as a transfer standard formeasurements at Reynolds numbers up to 4106 betweenNMIJ and PTB

    L. Cordova a,n, N. Furuichi b, T. Lederer a

    a Physikalisch-Technische Bundesanstalt, Germanyb National Institute of Advanced Industrial Science and Technology, Japan

    a r t i c l e i n f o

    Article history:Received 19 June 2014Received in revised form7 April 2015Accepted 19 April 2015Available online 24 April 2015

    Keywords:Interlaboratory comparisonUltrasonic ow meterReynolds number dependenceFlow traceabilityTransducer cavity

    a b s t r a c t

    The quality of any laboratory intercomparison depends to a large extent on the performance of the usedow meter. To nd a ow meter that is capable of reaching a reproducibility better than 0.05% requiresbounding all involved inuence quantities down to the required level. The present paper describes theefforts performed while qualifying a time-of-ight ultrasonic ow meter as a transfer standard. It wasdetermined that the most relevant inuence quantity besides the ow prole within the bulk ow is theeffect caused by the transducer pockets in the meter body. By taking advantage of a specially designedwindow chamber, it was possible to determine the magnitude of the errors introduced by the transducerpockets and to dene, based on the ndings, a procedure to perform a bilateral comparison between thehot water calibration facilities of the Physikalisch-Technische Bundesanstalt and the National Institute ofAdvanced Industrial Science and Technology. The results of the bilateral comparison are presented.& 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license

    (http://creativecommons.org/licenses/by/4.0/).

    1. Introduction and motivation

    Water is used as an energy transporting medium in every type ofpower plant involving turbines; also industrial and district heatingdepend on accurate measurements of ow rate. In most cases, theactual measurement uncertainty is in the order of 1%. Consequently,every improvement of the measurement uncertainties has directconsequences for the safety and efciency of the involved systems.

    Flow rate measurements in the eld are performed ideally byinstruments that have been tested at National Metrology Institutes(NMI) or at a calibration laboratory that has been accredited and/or is participating in prociency tests organized by the corre-sponding NMI as can be seen in Fig. 1. Any bias introduced by acalibration laboratory would have a direct impact on the price, onthe quality or on the competitiveness offered by its clients. Inorder for measurements to be globally consistent, it is requiredthat NMIs prove their mutual consistency periodically throughinternational comparisons. The Mutual Recognition Arrangementof the International Committee for Weights and Measures (CIPM-MRA) has established mechanisms in order to allow the NMIs toprove their mutual consistency transparently and based on thesame rules and principles. Actually there are more than 53 states

    and 152 institutes, designated by the signatory bodies, participat-ing in the CIPM-MRA.

    The traceability of a ow rate calibration facility is normallyassessed on a quantity-based calibration, i.e. mass, volume, time,density and temperature standards are calibrated separately. Only incases where there is a ow meter capable of delivering reproduc-ibilities much lower than the required calibration uncertainties it ispossible to provide a direct ow-rate traceability. This is possible inlow-ow hydrocarbon measurements as reported by Shimada. Highlyreproducible measurement instruments are available as seen, forexample, at the Calibration Intercomparison on Flow Meters forKerosene carried out on 1995 [10] and the CIPM-MRA internationalkey comparison of liquid hydrocarbon ow facilities CCM-FF-K2 [11].Without direct ow-rate traceability, systematic errors in any systemof the calibration rig might remain undetected.

    There are several relevant ow rate measurements in the eldperformed without a calibration as depicted in Fig. 1. This situation isgiven mostly in cases where the measurement conditions cannot bereproduced in a laboratory. Under these circumstances the only alter-native is to apply ow measurement technology that has a predictableworking principle that allows the use of similarity principles to infer thecalibration result and uncertainty of measurements under conditionsdifferent from those present during calibration.

    The relevant ranges for energy transport through hot water varymainly between 50 1C and 250 1C. Flow rates larger than 3500 m3/hhave been reported and Reynolds numbers up to 30106. According

    Contents lists available at ScienceDirect

    journal homepage: www.elsevier.com/locate/owmeasinst

    Flow Measurement and Instrumentation

    http://dx.doi.org/10.1016/j.owmeasinst.2015.04.0060955-5986/& 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

    n Corresponding author.E-mail address: [email protected] (L. Cordova).

    Flow Measurement and Instrumentation 45 (2015) 2842

  • to the CMC tables1 there is only one facility that comes close to theserequirements: the AIST, NMIJ (hereafter, NMIJ). With temperatures ofup to 70 1C and ow rates up to 12 000 m3/h, it is able to reachReynolds numbers up to 20106; the declared expanded uncertaintyvaries depending on the ow-rate range between 0.04% and 0.08%.The next ow rate facility that can be considered for hot water owtraceability studies is the heat meter testing facility of PTB. For adeclared 0.04% expanded uncertainty it is able to measure between4 1C and 90 1C and a ow rate up to 1000m3/h. Section 2 will givemore details on both facilities.

    In this sense, the ow measurement laboratories for hot waterof PTB and NMIJ cooperate in order to validate ow measurementprinciples that allow similarity conditions to be applied. And giventhat the required uncertainties to determine the inuence quan-tities acting on the ow measurement techniques are in the orderof magnitude of the uncertainties declared by the NMIs them-selves, PTB and NMIJ need to prove their mutual consistencybefore reliable experiments involving both laboratories are possi-ble. Steps towards this rst goal are described in this paper.

    Firstly, an overview on the used ow measurement technologyand on the calibration facilities of PTB and NMIJ is given. In thesecond part, the results of the characterization of an ultrasonicow meter made at PTB are shown in two steps: throughconventional linearity, repeatability and reproducibility tests usingan established industrial ow meter, and through the simulta-neous measurements of the ow prole and the ow meterindication at a very carefully constructed DN200 90D long testline using a specially designed window chamber. By using thecharacterization results, a strategy is dened to apply a robustindustrial ow meter as a transfer standard in less advantageousconditions. The transfer standard is provided with a tube bundle toincrease robustness against geometry differences in the inlet pipelayouts and internal pipe diameters. The nal part of this paperpresents the comparison results and provides rst conclusions onthe application of ultrasonic ow meters under conditions outsidethe calibration ranges.

    1.1. Traceability of ow meters outside calibration ranges

    An established ow metering technology based on the similar-ity laws concerns orice plate ow meters. They allow a bestpossible uncertainty, in the ideal case not smaller than 0.7% asextracted from ISO5167 [6], in any condition where calibration isnot possible. The basis for the ISO5167 is decades of enormousresearch efforts and ten thousands of internationally coordinatedexperiments.

    In the past few years, ultrasonic ow meter manufacturers havebeen introducing their products for applications where no calibra-tion is possible. Based on calibrations performed under laboratory

    conditions, they propose to extrapolate the uncertainty to levelsbelow 0.7% and replace differential pressure meters. Importantsteps towards global standardization of ultrasonic ow metertechnology have been undertaken in the GERG project on ultra-sonic gas ow meters [2].

    1.2. Ultrasonic ow meters

    The type of ultrasonic ow meter used most is the parallel pathtime-of-ight ow meter (hereinafter UFM). Its simplicity makes ita good candidate for the dened purpose.

    1.2.1. Ideal case integrationIn the ideal case, any path of a UFM installed at any position r=R

    when exposed to a fully developed ow prole shows a curvesimilar to the one depicted in Fig. 2(a). The area under the curverepresents the ow rate; when the bulk speed is dened to be one,the area under the curve is equal to the volume of a cylinder withunity radius and unity height (). Flow measurement through theUFM can be regarded as the problem of integrating the area underthis curve.

    If the ow is fully developed, any path can be used as a owmeter as can be seen in Fig. 2(b). 10 single normalized paths,referred to their own indication for Re 106, are shown as afunction of the Reynolds number. For every path position there is amonotonic relation between the indication and the real ow rate.

    The following equation describes the use of multiple paths Piand weights wi:

    Q kXn

    i 1wiPi 1

    The factor k of Eq. (1) is a correction factor of a semi-empirical natureintroduced to compensate for temperature and pressure variationsand to add empirical linearizing as seen, for example, in [12]. Theintroduction of a k-factor is comparable to the determination of thedischarge coefcient at orice plates. It would be desirable to nd avalid formulation for the UFM as is the case for orice plates asproposed by Reader-Harris et al. (as presented in [6]).

    1.2.2. Real case traceability limitsIt is easily concluded that the bias produced by the sum of any

    combination of parallel paths becomes asymptotic. In the idealcase, if the amount of paths n increases, the accuracy getsimproved. If the position of the nodes is selected based on aninterpolating integration technique, as the different forms of theGauss quadrature for example, more degrees of freedom areobtained making the method capable of compensating, to someextent, for small deformations on the projected ow prole causedby ow asymmetries. Several studies exist on this topic; see, forexample, [79].

    Considering the bulk ow within the ow meter, the ideal-ow-meter assumption requires that only axial velocity compo-nents are present. The existence of secondary components, radialor tangential, has a strong inuence and can produce errors in theorder of several percent. In the common case, where secondarycomponents can be considered to be constant while movingthrough the ow meter, if every path has a counter part down-stream with the opposite angle and at the same level, theintroduced error is cancelled out automatically. This conditionhas been taken advantage of by different ow meter producers.

    Considering the transducer pockets, they disturb the ow andintroduce secondary velocity components within and outside ofthem. Zheng et al. [13] determined numerically that the inuenceoriginated within the pockets is responsible for up to 4% of the totalsignal. This effect gets reduced at higher diameters where the

    Trac

    eabilit

    y

    Fig. 1. Traceability concept for hot water ow rate measurements. Representationused by Shimada [1] to show traceability on hydrocarbon measurements in Japan.

    1 Accessed on 02 June 2014 on http://kcdb.bipm.org/appendixc/

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 2842 29

  • transducers are negligible, compared to the total diameter: as statedin the recommendation of the PTC-18:2011 [3] for hydraulic turbines,the error introduced by protruding transducers for diameter of 1 m isin the order of 0.35%, for diameter of 5 m of 0.05%.

    The ISO 12242:2012 [4] and the AGA Report 9 [5] recommendassessing the reproducibility of an ultrasonic ow meter inreference to the calibration base line by testing the ow meterunder very adverse ow conditions. Measurements at differentpipe congurations known to produce strong secondary compo-nents and asymmetries are considered. It is expected that therepresented pipe layouts reect the worst conditions existing in areal application bounding the maximum errors that the instru-ments would produce. As shown by Drenthen et al. [21] or byCaldon [22], the introduced linearity errors are in the range of0.2%. This result provides a solid basis for interpolation, but ifextrapolation is required, more solid arguments are necessary.

    For the application of UFM in hot water measurements, it canbe assumed that given the low Mach numbers in the order of 0.01the path can be considered to be straight [14]. Time delaysintroduced, provided they remain constant, can also be neglected.

    2. The ow test rigs

    In the following section, both the PTB and the NMIJ facilities arepresented. Special attentionwill be given to the calibration facility of PTBthat was used for the characterization of the ultrasonic ow meters.

    2.1. Flow rate facility NMIJ

    The ow test facility of NMIJ has been described in detail in theprevious publications [15]. The ow test facility of NMIJ is based onseveral weighing systems working at ambient temperature. In orderto make the higher temperatures traceable, it is required to transferthe accuracy obtained by the gravimetric systems, to a temperedvolumetric system. The facility used for the measurements pre-sented in this paper is the prover system shown in Fig. 3. Thisfacility generates ow rates from 200 m3/h up to 800 m3/h at 20 1Cup to 80 1C 70.5 1C. The prover system is a core component toprovide traceability to the large Reynolds number facility. Thehighest pressure of the test line is 0.7 MPa. The nominal pipediameter of the test line is DN200 and the length of the test line isapproximately 12 m. The maximum Reynolds number in the testsection is approximately 3.7106. The reference ow rate is givenby the volumetric method of the prover. Inside of the pipe, there is aspherical ball with a diameter about 2% larger than the pipe

    diameter to avoid leakage. The ball activates the start and stopdetection sensors whenmoving from one side to the other. The owrate is given as the standard volume between two detection sensorsdivided by the elapsed time. The standard volume between the twosensors is calibrated by the gravimetric system through the transferow meters. The uncertainty sources of the prover system are thestandard volume of the prover, correction of the standard volumefor the temperature and pressure, and the measurement of theelapsed time. As mentioned, the standard volume is calibratedusing the gravimetric system and the transfer meter, and it is thedominant uncertainty source of the prover system. The expandeduncertainty (k 2) of the facility is 0.068%. The minimum elapsedtime is 15 s. The measurement is normally repeated 20 times.

    2.2. Flow rate facility PTB

    The heat meter testing facility of PTB (Waermezaehlerpruef-strecke WZP) is a gravimetric ow test rig for temperatures up to90 1C. A more detailed description is available at [16]. A schematicof the facility is shown in Fig. 4. It is divided basically into twolevels: the basement level with the ow rate generation systems,and the upper level with the test lines and the measurementsystems. The ow rate is generated with two sets of pumpcascades, with an overow constant pressure tank in between toensure highest ow rate stability. Since the measurements areperformed on a ying start/stop basis, a diverter system has to beused. Evaporation at higher water temperatures is controlled byreducing the vapor concentration gradient in the air near all freewater surfaces. This is accomplished by encapsulating the divert-ing system and by introducing saturated tempered humid air intothe empty tank before the measurements. Evaporation cannot becompletely avoided: thus by performing a water vapor massbalance based on humidity measurements on the air evacuatedby the water, the amount of water loss can be estimated.A weighing scale calibrated on a daily basis is the referencesystem. The heat meter testing facility of PTB is designed, main-tained and used to deliver an expanded ow rate realizationuncertainty not larger than 0.04% and a very high repeatabilityfor temperatures between 4 1C and 90 1C and ow rates up to1000 m3/h. The length of the test lines is 25 m.

    During 2013, the most important components of the owcalibration facility were overhauled. After more than 100 000diverter motions, the diverter systems were renewed. The forceisolating and force transmitting components of the weighing scalewere adjusted and a redundant strain gauge system was alsoinstalled. The humidity determination system was improved. Given

    Fig. 2. Theoretical path indication as a function of the Reynolds number. The ow prole projection was normalized to the bulk speed based on the semi-empirical modelproposed by Gersten. A biased representation allows for better comparison. (a) Flow prole projection as seen by ultrasonic ow meter for Re 2 106. (b) Projection ofsingle paths for different Reynolds numbers referred or biased to Re 1 106.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 284230

  • these important hardware changes, a comprehensive characteriza-tion was required. A more detailed publication of the results is thesubject of a different paper. Here, only an overview will be given.

    The uncertainty of the ow rate facility is assessed, divided intofour groups: the mass measurement, the density measurement,the timing error, and the process-related components. The lastgroup includes all additional mass correction due to thermalexpansion, air entrapment, buoyancy variations and evaporation.

    Fig. 5(a) shows graphically how the uncertainty componentsinteract, in this case when an expanded uncertainty of 0.04% is

    required. We can see that if larger ow rates are required, themajor component is the timing error; for lower ow rates theprocess-related corrections have the largest contribution.

    Fig. 5(b) shows the inuence of temperature and lling volumeon uncertainty for different lling volumes. The resulting uncer-tainties from 0.025% up to 0.05% are shown. It can also be seen thathigher temperatures play only a role for lower ow rates. Thelargest problem at higher temperatures is evaporation. There areseveral measures applied in order to compensate for or to avoidevaporation. It has been determined empirically that these mea-sures are less effective at lower ow rates. The uncertaintiesshown in Fig. 5(b) do not include the contribution of the owmeter under test.

    2.3. Internal consistency test for the ow calibration facility

    The only component of a gravimetric ow rate facility thatdepends on the Reynolds number is the diverter; the reason is theow prole at its entrance. Depending on the ow rate and on thetemperature, the ow prole will change. This effect is systematicand is overlapped with the temperature dependence of thepneumatic actuator system. Consequently, the timing error isdetermined periodically at different temperatures and ow rates.If the corrections are applied correctly, there is no residualReynolds number dependency left on the measurement resultsof the WZP. By taking this into account, when calibrating an oriceplate that has a strong Reynolds number dependency at differenttemperatures and ow rates, it should be possible to determine ifthe different components of the gravimetric ow rate facility areworking properly. This was done with a highly repeatable DN200orice plate with 0:75. The results are shown in Fig. 6.

    Fig. 6 shows six different temperatures where the ow rates390 m3/h, 475 m3/h, 595 m3/h and 745 m3/h have been measuredrepeatedly. The results are presented as a function of the Reynoldsnumber Re v D=T, where v is the bulk velocity, D the pipediameter and the kinematic viscosity that is a function of thetemperature. Given that a single discharge coefcient can berealized at different conditions, the results of measurements ofthe discharge coefcient at different temperatures overlap as seenin Fig. 6, i.e. the rst point at 745 m3/h corresponds to 390 m3/h ata different temperature. For these two ow rates different llingtimes were used (156 s and 82 s). If there is any time dependent

    Fig. 3. Test facility with prover system of the NMIJ.

    Fig. 4. Operational area and basement of PTB heat meter ow test rig (WZP).

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 2842 31

  • error, it should be visible at these points, but no differences weredetected.

    The error bars shown correspond to 0.01% of the dischargecoefcient; the black curve is the result of a regression based onthe ReaderHarrisGallagher equation as presented in [6]. As canbe seen, there is extraordinary agreement across all ow rates andtemperatures. The pure Reynolds number dependency of theorice plate is observed. Therefore, we can conrm that allsystems are working properly and that all corrections are beingapplied correctly. Additionally, the observed results also conrmthat the ow prole at the WZP is the same for coincidentReynolds numbers at different temperatures and ow rates, sinceorice plates are very sensitive to ow prole changes.

    3. Methods

    Two ow meters are used for the experiments. An industrialow meter as a transfer standard and a specially designed owmeter with an optical access or window to the body to performprole measurements. Initially, preliminary measurements areperformed in order to dene the performance of the transferstandard and to dene the best conditions to perform thecomparison. The next stage is to characterize the ow prolewithin the ow meter at the calibration facility in order to be ableto dene in a next step the ideal working conditions and the main

    inuence quantities for using an UFM, but applying both time-of-ight and also ow prole measurements.

    3.1. Preliminary measurements

    The industrial ve-path ow meter (I-UFM) used is a part of ameter run package composed of a 4 m long upstream pipe and a1 m downstream pipe. To guarantee repeatable measurementconditions and robustness against differences in the upstreamow proles an ISO5167 tube bundle ow straightener (TB) wasinstalled. This is necessary because even if the involved facilitieshave long upstream pipes the internal diameter sizes do not matchexactly. In addition, the upstream section anges are pinned toguarantee repeatable mounting. For the analysis only raw datadelivered by the I-UFM were used. All correction and compensa-tion factors provided by the manufacturer were deactivated,because during characterization of the different inuence quan-tities, any overlapping correction would disturb the analysis.

    As a rst assessment, linearity, repeatability and reproducibilitytests were performed. All results were in a band of 70.1% for agiven conguration, but no clear Reynolds dependency could beobserved, as explained in Section 1.2.2. Repeatability reachedvalues in the range of 0.02% and 0.04% for all temperatures, owrates and congurations. Regarding reproducibility, measurementswith and without TB differed by about 0.4%. A surprising resultwas obtained by changing the exact position of the TB. Differentrotation positions produced differences of about 0.15%. Theseresults are shown in Fig. 7.

    To test robustness against ange mismatch, the upstreamsection was mounted with a 0.5 mm off-axis on its upstream side.The produced differences were systematically in the rangeof 0.05%.

    The preliminary tests in summary:

    Independent of the measurement conditions the I-UFM deli-vers a highly repeatable result.

    If measurements with a reproducibility better than 0.1% arerequired, UFMs should be mounted with great care in terms ofalignment and conguration

    The TB, in spite of fullling the requirements of ISO5167,introduces repeatable asymmetries that prevail after the 4 mupstream pipe and depend on its rotation angle.

    Since the geometry and location of the transducer pockets vary,the unknown systematic effects causing the errors might bedifferent for each single path. If besides the axial velocity relatedpath velocities Pi each measurement path is inuenced by the

    Fig. 5. (a) Uncertainty contribution factors for the required expanded uncertainty of 0.04% in percent for 80 1C. (b) Different expanded uncertainty values that can be reachedat 50 1C or 80 1C for different lling volumes.

    Fig. 6. Discharge coefcient as a function of the Reynolds number for an oriceplate at PTB for a 0:75 DN200. The error bars correspond to 0.01% of thedischarge coefcient.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 284232

  • error ei, the following equation would apply:

    Q kXn

    i 1wiPi

    Xn

    i 1wiei 2

    The errors introduced by each path are unknown; in order tominimize the total error, it might be necessary for the weighingfactors wi of Eq. (2) to acquire also negative values. If an additionalsummation term Qk is added instead, as in Q k

    Pni 0 wiPiQk,

    the problem of using negative weighing factors can be avoided.However, the determination of both types of corrections is at thecurrent state of the art only possible through empirical treatment.By determining the calibration factors at the same laboratorieswhere the UFM is tested, unpredictable correlations would beintroduced leading to a biased estimation of consistency.

    Even if all wi and k of Eq. (2) are assumed to be known for anideal case, since the distribution of ei cannot be guaranteed to berandom, there will be systematic inuences that are not elimi-nated through averaging that invalidate the obtained results.

    Therefore we decided to use, instead of the weighed summa-tion of all single paths, each path independently, free of anyempirically determined constants. Due to its symmetry and tothe maximum length, the central path is predestined to serve as areference.

    Only by knowing the ow prole within the UFM will it bepossible to determine the performance of the ow meter. To makethis possible, a hybrid ow meter with an optical access has beenspecially built. The design goal was to enable velocity prolemeasurements within the UFM by means of Laser DopplerVelocimetry (LDV) and Ultrasonic Velocity Proling (UVP) but

    without introducing additional disturbances. UVP offers theadvantage of measuring secondary components if they aremounted on the same plane as LDV. Refer to [26] for furtherdetails.

    3.2. Velocity prole measurement

    3.2.1. The window chamberThe designed window chamber (WCH) is based on a 5-path

    UFM (UFM-WCH). The outer paths P1 and P5 are on the samevertical plane mounted at 451 from the ow axis; the three centralpaths P2, P3 and P4 are on a plane at 451 space from the owaxis, perpendicular to the outer path plane. Normally, P3 ismounted on the same plane as P1 and P5, but by changing itsposition as seen in Fig. 8(b), there is enough space left forpositioning an LDV and UVP access in between the paths. Fig. 8(b) shows the glass insert mounted on the UFM-WCH and thetransducer pockets of paths P2, P3 and P4. There are four insertsmounted in total every 901. The glass insert was thermallyhardened and polished afterward to minimize any gaps or dis-turbances on the wall. Hardening had a negative inuence on theoptical quality, but it was unavoidable in order to guaranteeoperation safety. Due to the large surface of the insert exposedto the internal pressure, forces of several thousand Newton areapplied. These forces could cause small changes to the thickness ofthe sealings, which would have negative consequences on thebeam positioning. Therefore, an elaborate sealing system has beendesigned to avoid displacement of the glass due to geometricalvariation of the seals, but without compromising safety. Formaking the UVP measurements the inserts have been nishedusing polyoxymethylene.

    The setup for the LDV measurements can be seen in Fig. 10. Aregular XY traversing system for the LDV probe would only providesmall optical access into the ow. Therefore, a combination of acircular shaped traverse and a linear table has been designed. Bypositioning the center of the circular shaped traverse near theinsert, a much larger view of the ow is possible as seen in Fig. 9.The gure shows the typical standard deviation of the mean axialspeed obtained, estimated from the empirical measured turbu-lence and the amount of valid bursts detected. As can be seen, theamount of burst varied greatly, such that uz shows values up to2.5%, the reason for this was the numerous reections comingfrom the stainless-steel body, and the poor optical quality ofthermally strengthened glass.

    The uncertainty of the LDV measurements was determined tobe under 1% for a single point. This has been accomplishedthrough the characterization of the traversing system at a coordi-nate measuring table and by means of a rotating disc calibrationfor the LDV probe. More details on LDV calibration uncertainty ofthe equipment used can be found in Ref. [17].

    Fig. 7. Measurements without TB and with the TB mounted in 3 different rotationpositions. The conditions were 20 1C and 390 m3/h. The single measurement pointsare shown to give an impression of the repeatability.

    Fig. 8. Window chamber details and a tube bundle frontal view. (a) DN200 window chamber. (b) Inner view of the window chamber. (c) Tube bundle.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 2842 33

  • 3.2.2. The ow proleThe main purposes of LDV were to determine the reason for the

    large differences present under the different installation conditions

    and to conrm that a fully developed prole exists. The resultingprole is shown in Fig. 12(a).

    The window chamber was positioned with an upstream lengthof 90D. The conguration is shown in Fig. 11. A honeycomb-typeow straightener with square cells (FS) and a perforated plate owconditioner (FC) were installed.

    The task force for Laseroptical Flow Diagnostics (TFLD) basedon the work of Yeh and Mattingly [18] recommended the use offour performance indicators for ow calibration facilities for heatmeters. Three of them will be used here: indicators for the owprole peakedness, for the ow prole asymmetry and for theturbulence intensity. The ow prole peakedness and the owprole asymmetry are dened for diametral (2D) slices of the owprole. The turbulence degree is dened for the central core of theow. The Guidelines for the uid mechanical validation ofcalibration test-benches in the framework of EN-1434 [20] givea full description on its calculation and establishes limits for anearly fully developed ow prole.

    The ow prole peakedness and asymmetry indicators aredened for the axial component of the velocity, and are calculatedaccording to the recommendation of Yeh.2 To enable the compar-ison of indicators across different ow rates and pipe sizes, Yehnormalized the results to a fully developed ow prole. The FTLDrecommended to assume as a fully developed ow prole thesemi-analytical formulation proposed by Gersten [19] for smoothpipes. See [20,18,19] for further details.

    Since the view of the ow prole is limited, the performanceindicators will be calculated for r=Rr70:65, which is the largestcoaxial circle that can be fully measured. Consequently, given the factthat the indicators for peakedness and asymmetry are dened for 2Dslices with limits r=R 71, the estimations for r=Rr70:65 will bebiased. In case of the prole factor, owprole changes in the central partof a ow prole are overrated if seen only in a 2D slice, since only thelength is used as a weight instead of the area, as in the real case. But inorder to allow comparisons and rating according to the recommendationof the TFLD, 2D slices will still be used for the calculations. The relevantperformance indicators are summarized in Table 1.

    Fig. 12(b) shows measurements at 600 m3/h using LDV andUVP. Both systems deliver the same results and are very close tothe theoretical prole. The differences encountered are withintheir declared uncertainties. The LDV and UVP measurementswere performed with the parameters shown in Table 2.

    As can be seen, the ow prole at the position of the UFM-WCHcan be considered to be fully developed. The LDV measurementswere performed from both sides. This was achieved on differentdays and also after taking out and remounting the UFM-WCH. Theresults are consistent and conrm the reproducibility of theconguration.

    3.2.3. Measurements of the TB proleThe following measurements were obtained with the TB

    installed 20D from the UFM-WCH.At the 1201 position shown in Fig. 13(b) some measurement

    points delivered less than 100 bursts for the established time. Forthis reason it was not possible to calculate some of the perfor-mance indicators reliably.

    It was assumed initially that the ow prole should rotatetogether with the TB, but as can be seen in Fig. 13 the prole doesnot rotate, it changes every time. It seems that small asymmetrieson the anges and on the TB cause the conguration to be a littledifferent for each position of the TB.

    The maximum speed on the TB ow prole is about 4% higherthan on the ow prole measurement without TB. This is clearlyseen in the prole factor values; the undisturbed ow has a prolefactor of 0.94, while the measurements with TB about 1.2. If theow rate were measured only on one diametral path, it would beexpected that the ow rate is overestimated due to the peak in thecentral region. But the opposite case is observed: an underestima-tion of about 0.4% was measured. This is an indication that thepeak is not the only reason for the differences. Either the inuenceof the TB on the transducer pockets, or undetected secondarycomponents, or both, are causing the bias.

    The position 01 is apparently the best choice to install the TB.The error is small and the prole has the most symmetric shapeconsidering the maximum Ka, if only the central diametral Ka

    0 isconsidered. The 2401 position seems to be symmetrical, but it canbe seen that the peak position and the gravity center have a largerdisplacement from the axis.

    Fig. 9. LDV measurement grid and standard deviation of the mean axial speed uz .

    Fig. 10. Typical set-up for an LDV measurement using the UFM-WCH.

    Fig. 11. Installation conditions for the measurements of the ow prole. FS is theow straightener and FC is the ow conditioner.

    2 Yeh introduced different performance indicators to evaluate the inuence ofa reducer installed in front of an orice plate using LDV. Among the family ofindicators he proposed in [18], P5 and S10 are the basis for the work of the TFLD.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 284234

  • 3.2.4. Radial components on the central pathThe central path is insensitive to swirl, provided that the swirl

    is coaxial with the pipe axis. Recalling that the UFM-WCH isinstalled behind a ow straightener installed 90D upstream of themeasurement position, we can assume that the secondary com-ponents within the pipe diameter are negligible. But in the regionnear the transducer, the pocket might introduce an additional bias.UVP has been used to determine the magnitude of these effects.Fig. 14(a) shows the measurements using a 1 MHz UVP with a13 mm effective diameter. The pulse repetition frequency was1805 Hz and the resolution 0.005 m/s.

    To interpret the results of Fig. 14(a), Fig. 14(b) has to beconsidered rst. This gure shows the shape of the UVP measure-ment volume within the pipe and inside of the transducer pocket.In contrast to LDV which provides a good spatial resolution, theUVP measurement volume considers the speed of a much largerarea and is affected by reections. For the bulk ow within thepipe, the spatial resolution is small enough, but for small scaledmeasurements as is the case with the transducer pocket, only avery rough idea of the ow prole can be given.

    In addition, when the UVP measurement volume is truncatedby the pipe wall, reections occur deforming its space. Special carehas to be taken if these effects are expected. Signals originatedfrom reections can be ltered out by limiting the receiving timewindow. In some cases, reected doppler signals are weakcompared to the signals coming from the main ow and can beneglected, but since the size of the volume left outside of the wallis reduced a displacement of the effective center has to beconsidered. In our case, in order to be able to receive signals frominside of the cavity, the time window has not been reduced.Reections cannot be neglected and the shape of the measure-ment volume is affected.

    The shapes of the measurement volume for different depthsinuenced by reections are shown in Fig. 14(b). The shapes have beensimplied assuming that the pocket is squared. In the real case, only thefront face of the transducer is at, as in the squared case producing astronger signal than the pocket wall. The section EE of Fig. 14(c) isdepicted in Fig. 14(b) (the transducer face is located on the upper side).

    Table 1Symbols used in the performance evaluation gures.

    Symbol Limitsa Units Description

    D mm Pipe diameter 208 mmRe Pipe Reynolds numberQ0:65 % of Q Flow rate within r=Rr70:65Ka % of D The asymmetry factor for r=Rr70:65Ka

    0 o1 % of D The asymmetry factor for the horizontal path for r=Rr71Kp The maximum prole factor for r=Rr70:65Kp

    0 0.8 to 1.3 The prole factor for the horizontal path for r=Rr71Tu o2 Turbulence factordp mm Distance from the peak to the pipe axisdv mm Distance from the gravity center of the ow to the pipe axisumz=uo Maximum relative axial uid velocityumz m/s Maximum axial uid velocity

    a Extracted from [20].

    Table 2LDV and UVP main specications.

    Property LDVa UVPb

    Velocity resolution 0.005 m/sAverage meas. Volume width 1 mm 25 mmAverage length 5 mm 3 mmAverage height 1 mm 25 mmNumber of points 100 224Time per path 60 min o1 minTracer particles 10 m 100 m

    a 75 mW Nd:YAG 532 nm and 45 mm beam distance and 250 mm focal length.b MET-ow UVP-DUO and 1 MHz transducer at 71. The parameters vary

    depending on the requirements. The data serves only as a reference and corre-sponds to the results shown in Fig. 12(b).

    Fig. 12. DN200 LDV and UVP measurements after 90D upstream pipe. (a) Fully developed ow showing performance indicators with LDV measurements from both sides.(b) LDV and UVP velocity measurements at DN200 and 600 m3/h.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 2842 35

  • Due to the deformation of the measurement volume we canassume that measurements without interaction of the wall arecorrect, i.e. up to a depth of 208 mm. In the near wall region, theradial component seems to be dependent on the ow rate. Forthe region within the pocket, the rough spatial resolution does notallow drawing nal conclusions on the ow prole within thepocket. The U-shaped measurement volumes might be simulta-neously perceiving radial and axial components. This could explainthe two peaks found in Fig. 14(a) at 213 mm and 222 mm. If thepeaks were only caused by radial components, the peak at222 mm would indicate a ow rate leaving the bottom of thecavity, which cannot be true. Therefore we can assume that thepeak is caused by the axial components of a vortex in the cavity.

    We can conclude from this experiment that even for the simplyshaped central transducer pocket, the inuences on the main owcannot be ignored. Even if a fully developed ow free of secondarycomponents is given, the transducer pockets interact with the bulkow causing radial components to occur.

    The study of the ow within UFM cavities is a complex problem.For a qualitative impression see the eddies which formed in two non-diametral pockets in Fig. 15 at 390 m3/h. Air bubbles were introducedto make the eddies visible with the simple eye. Microbubbles used forUVP are not visible. The center of rotation of the vortex coincides withthe axis of the ultrasonic path. This is relevant for UFM since mostcomponents remain unperceived, but up to what extent the eddyinuences the ow outside the pocket is an actual topic of research.

    Fig. 13. Tube bundle ow prole at 3 different positions for 390 m3/h and 30 1C. (a) Position 01, (b) position 1201, and (c) position 2401.

    Fig. 14. UVP central path transducer pocket measurement. (a) UVP velocity measurements on the UFM-WCH for the central pocket. (b) Divergence of the 1 MHz ultrasonicbeam and shape of the measurement volume at different depths. (c) Pocket shape and vortex scheme.

    Fig. 15. Qualitative indication of the eddies existing within the pockets with 30 1C and 400 m3/h for r=R 0:8 on the left and r=R 0:5 on the right. The ow direction is fromleft to right.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 284236

  • PTB is actually using the capabilities of the window chamber tocharacterize the ow within ultrasonic in-line ow meters. Forthis purpose, differently shaped cavities will be installed andcharacterized by means of LDV and UVP. The cavity characteriza-tion project is in its initial stage.

    3.3. Single path measurements

    The next step is to test the performance of the UFM-WCHunder fully developed ow conditions. For this purpose measure-ments were performed at temperatures 20 1C, 30 1C, 40 1C and50 1C and at the ow rates 390 m3/h, 475 m3/h, 595 m3/h and745 m3/h. Since all corrections have been turned off, the tempera-ture correction was compensated subsequently according tokT 1t3 13t.

    The result of each normalized path at each temperature andow rate condition is depicted in Fig. 16. It is difcult to recognizedeviations on the ow prole based on this gure alone. For amore detailed view, the relative deviation of each path relative tothe ideal, fully developed ow prole at the respective Reynoldsnumber is shown in Fig. 16(b).

    For a fully developed ow all points should be around the 0-line. But in this case, the maximum difference to ideal conditions isabout 4.4%. The required geometrical displacement to producesuch a large error is about 2 mm. Manufacturing tolerances can beguaranteed to be far below 0.1 mm. Therefore this deviation can beattributed to the actual existing ow prole. It is also remarkablethat the results of symmetrically mounted paths are not symme-trical. On a large scale, the asymmetry is independent of thetemperature, of the ow rate and of the Reynolds number sincethe rough position of the path errors remains constant. Todetermine if there is some dependency on a smaller scale, everysingle path curve will have to be observed independently.

    Fig. 17 shows the results of the measurement campaigns. Fig. 17(a)(e) shows the relative error of each path at different tempera-tures and ow rates as a function of the Reynolds number. Eachpath has been considered as an independent ow meter scaled tot between 0.4% and 0.6% using a different proportional factorfor each of them. The error that these different factors would haveon a fully developed ow prole is shown in Fig. 17(g). Theweighted sum of the single paths is shown in Fig. 17(f).

    A rst look reveals immediately that the Reynolds numberdependency is given only for path 3. Apparently paths 1 and5 have no Reynolds number dependency, but rather a ow ratedependency since independent of temperature, the maximumow rates behave similarly. Measurements on path 5 were invalid

    for 40 1C and 50 1C. But even only for 20 1C and 30 1C it can beobserved that the dependency is given rather for the ow rate.

    Paths 2 and 4 are distributed in a narrower band. The shapes ofthe curves are also rather independent of the Reynolds number,but a clear dependency on the ow rate can be disregarded due tothe results for 50 1C.

    Path 3 delivers a strong dependence on the Reynolds numberas expected with a range of about 0.7%. But as can be extractedfrom Fig. 17(g) the theoretical curve has a different slope and theconsidered range has a slope of 1.2. Given that LDV measurementshave measured the central path completely and proved a nearlyfully developed ow condition, and considering also that UVP hasproven that no considerable disturbances are present on the wallto wall measurements, the large differences in the steepness andin the position of the curves for the central path can be clearlyattributed to the inuence of the pockets.

    The integration capability for removing disturbances is remark-able. This can be seen in Fig. 17(f). Considering the deviationsencountered on each path, the nal result is very at and within anarrow band. The absence of signals on path 5 for 40 1C and 50 1Chas been compensated automatically with the internal algorithmsof the ultrasonic systems installed.

    For the purpose of a bilateral comparison and for the validationof the owmeasurement principle, it is not enough to consider theresults of Fig. 17(f), since the reasons for the deviations on eachpath are not understood. Nevertheless, the results given by path3 conrm the potential of this technology to be capable of servingas high quality transfer standard and of providing a solid basis forextrapolation.

    3.4. Design of the comparison

    Each path will be used as an independent ow meter. But onlythe central path will be used as a reference. The indication of theouter paths will serve as an indication that the ow conditions atboth laboratories are the same and constant. The weighted sum ofthe ow rate will be considered only as an initial indication.

    In order to be able to detect possible ow rate and temperaturedependencies, the measurement points will be chosen in such away that constant temperatures, constant ow rates, but alsoconstant Reynolds numbers will be aimed at whenever possible.

    The industrial ow meter used has an internal diameter of202.7 mm; the upstream and downstream pipes have a diameterof 206 mm. In order to avoid a step on the wall, the I-UFM has asmall conical reduction. Given this change in the geometry, a fullydeveloped ow will never be given. If we consider also that the

    Fig. 16. Normalized path speeds on a ow prole projection of the UFM-WCH and its relative path errors referred to a fully developed ow prole. (a) Flow prole projectionfor a normalized bulk speed of 1 and single path results for 4 ow rates and 4 temperatures. (b) Single path speed normalized to fully developed ow conditions. Meanvalues for 4 ow rates and 4 temperatures are shown at each position. (c) UFM path conguration scheme. For the UFM-WCH, P3 is parallel to P2 and P4; for the IUFM it isparallel to P1 and P5.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 2842 37

  • Fig. 17. Measurement results for the UFM-WCH at PTB. (a) Path 1, (b) path 5, (c) path 2, (d) path 4, (e) path 3, (f) sum, and (g) expected errors.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 284238

  • pipe diameters upstream of the meter run package in bothlaboratories are different, no dened conditions would be possible.For this reason, as mentioned in Section 3.1, in order to be moreindependent of the installation conditions, it has been decided toperform the measurements using the TB. The rotation position atthe 01 was xed for the measurements, since it gives repeatableand the most symmetrical results. As seen in Fig. 18(b), the TB isalways installed 20D in front of the ow meter.

    4. Results

    The results are presented in two parts: the single path resultspresented as a ow prole indicator and the path by path relativeow rate error results.

    4.1. Measurement conditions

    The measurement results have been obtained within 2012 and2013. The chosen measurement points shown in Fig. 18(a) enablemeasurements at constant ow rates, constant temperatures andnearly constant Reynolds numbers. The Reynolds numbers are notexactly the same, but they are close enough to allow a Reynoldsdependency analysis. Each point was repeated at least 5 times atPTB and 20 times at NMIJ.

    The pressure was held at both laboratories at 3 bar. The pipingconguration is shown in Fig. 18(b). Fig. 16(c) schematizes theultrasonic path conguration of the used I-UFM.

    4.2. Path projection results

    The obtained relative path errors are shown in Fig. 19. Thesingle errors are connected with lines to improve readability. As inthe case of the UFM-WCH, the single path errors are as much as 5%distant to the 0-line.

    The dispersion of the different points is the highest for theouter paths P1 and P5 (at r=R 70:8) and is reduced for P2 andP4 (at r=R 70:5). The dispersion of P3 is in most cases thelowest. In the case of PTB, the measurements at 67 1C and 80 1Chave a stronger dispersion for P3.

    Apparently, the ow conditions at NMIJ vary depending on thetemperature. If closer attention is paid to P4 and P2 we can seethat while P4 increases with rising temperature, P2 is reduced. Thesame can be observed at paths P1 and P5 but to a lesser extent.

    The only reason for this kind of disturbance is swirl. But howcan swirl be generated at NMIJ and not at PTB if the sameconguration were being used? A TB is introduced to eliminate

    swirl coming from the ow test rig. Therefore, it can be assumedthat if swirl is the cause for the path asymmetry, it was generatedby the tube bundle itself, but only in the conguration at NMIJ,since the measurements at PTB have been proven to be swirl-free.Similar experiences with TB have been made by Brown et al. [24].

    Considering the measurements at 20 1C and at 80 1C of P4, theasymmetry has doubled from about 1% to 2%. If the TB generatesswirl 20D upstream of the I-UFM, a decay as a function of thedistance and of the Reynolds number should exist. Referring to theexperimental results of Mattingly et al. [23] for the maximumswirl angle, the decay 20D downstream of the TB should vary verylittle between 61% and 64% for the considered Reynolds numberrange. This would suggest an apparent independence of swirl tothe Reynolds number. But when the temperature and, conse-quently, the Reynolds number changes, the swirl effects changeremarkably, which is in contradiction to the ndings of Mattingly.The last possible reason for swirl would be a temperaturedependent change in the pipe and ange geometry due to thermalexpansion of the solid components affecting the tube bundle itself,or the supporting system of the pipe setup.

    In any case, the path error asymmetry is caused by thedisturbances in the pockets, by swirl or by an interaction of both.Fortunately, P3 is not affected by the observed effect.

    4.3. Path by path comparison results

    Fig. 20 shows all the results of the measurements at NMIJ andPTB. Fig. 20(a)(e) shows the relative error of each path consideredas an independent ow meter. In order to make the resultscomparable, a different proportional factor was used with eachpath. The effect on the relative error that these used factors wouldhave on a fully developed ow prole is shown in Fig. 20(g). Theweighted sum of the single paths is shown in Fig. 20(f).

    The differences in the outer paths between both laboratoriesbecome evident. P1 shows differences of up to 1% for the 67 1C and80 1C measurements. The lower temperature seems to be in betteragreement. The results of NMIJ are widespread in contrast to theresults of PTB which show a more consistent behavior in terms ofthe Reynolds number. A direct ow rate dependency seems toaffect the results of NMIJ. The error increases for the lowest owrates and decreases for the higher ow rates. P5 shows a clearReynolds dependency for both laboratories; however, the differ-ences are between 0.2% and 0.4%.

    P2 and P4 show for PTB a consistent Reynolds dependency. ForNMIJ the paths P4 and P2 but to a lesser extent show thetemperature dependent error. As in the case of P1 and P5, theerrors are always in opposite direction.

    Fig. 18. Measurement conditions for the measurements at NMIJ and PTB using the I-UFM. (a) Preferred measurement points. (b) Installation conguration at PTB and atNMIJ. The TB is installed in both cases about 20D in front of the ow meter.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 2842 39

  • As expected, we can see that the error curve of P3 is consistentfor both laboratories. A clear Reynolds dependency is observed butsome additional temperature effects are observed; a difference ofup to 0.3% exists at Re 2 106. By taking a closer look at the dataof P3, we can observe that the lines for 20 1C and 40 1C are in fullagreement, while for 50 1C, 67 1C and 80 1C the differences rise upto 0.15%.

    The fact that the errors occur in different directions for eachcouple of symmetrical paths has taken advantage of when theweighted sum is used as a ow rate indication. But in contrast toP3, the measurements at 20 1C show a larger difference.

    Fig. 20(g) summarizes the results of all other curves. Each pointrepresents the root mean square differences of each temperatureacross all ow rates. The gure offers an overview of theperformance of each path. The differences are smallest for P3and for the weighted average, the latter always below 0.1%. For allother results 0.2% and more can be expected.

    5. Discussion

    The preliminary measurements using the WCH at the carefullyconstructed 90D step-free and gap-free honed upstream pipe haveproven that even if nearly fully developed conditions exist, everysingle path introduces an additional error. The magnitude of theerror is up to 5%. If measurements are performed with and withoutow conditioner, differences of about 0.4% can be observed if theweighted sum indication is used. These differences cannot beattributed to the peakedness of the ow prole introduced by thetube bundle, since peakedness would produce deviations inthe opposite directions; the cause of the differences is probablythe transducer pockets.

    The velocity eld within the transducer pockets has beenassessed qualitatively with the WCH. The pockets with r=R 0:5and r=R 0:8 have eddies that are coaxial with the transducer; thecentral pocket has an eddy whose axis is perpendicular to thetransducer axis. The inuence of the central pocket has been

    estimated via UVP as seen in Fig. 14(a); if the ow rate isincremented, the inuence is also increased. P3 is the only paththat shows a clear Reynolds dependency and, due to its position, itis insensitive to symmetrical swirl. Because of this, it is assumedthat the introduced error of the central pocket is also dependenton the Reynolds number. The same condition cannot be applied tothe outer paths. There is not enough knowledge to explain theshape of the error curves. Therefore, actually only the central pathis capable of serving as a transfer standard.

    Weighted summation is a robust method to deal with distur-bances, if used adequately the UFM will deliver results within0.15% . But the weighted summation does not only have positiveaspects. If the measurement results with P3 at 20 1C from Fig. 20(e) are considered it would be expected that the weightedsummation also delivers a good result, but an error in the rangeof 0.09% is introduced.

    In order to prove mutual consistency between the two labora-tories, a transfer standard with a reproducibility at least in theorder of their declared uncertainties should be used. UFMs haverepeatabilities in the range of 0.02%. Their reproducibility dependsin theory mostly on the ability to establish the same ow prole.Consequently, the measurements can be considered valid if thesame ow prole is present. In the case of Fig. 20(e), we canobserve that measurements at 20 1C and 40 1C follow exactly thesame pattern.

    The only cause of overlapping results in spite of having adifferent ow prole would be the existence of the same bias atboth ow test rigs. But since PTB is using a gravimetric systemwith a lling volume of 17 m3, and NMIJ is using a completelydifferent measurement principle with a volume of 3.5 m3, theprobability that a possible error introduced by the ow prole anda hypothetical bias of the ow test rigs is fully compensated fortwo temperatures and ve ow rates is negligible.

    Consequently, we consider as conrmed that PTB and NMIJ areconsistent for 20 1C and 40 1C and ow rates up to 740 m3/h. In thecase of PTB, the measurements with the orice plate shown inFig. 6 show that there is no reason to believe that only

    Fig. 19. Prole projection for the measurements using the I-UFM at PTB and NMIJ. P1 and P5 correspond to r=R 70:8, P2 and P4 correspond to r=R 70:5 and P3corresponds to r/R0. (a) 20 1C, (b) 40 1C, (c) 53 1C, (d) 67 1C, and (e) 80 1C.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 284240

  • measurements at 20 1C and 40 1C are correct. Therefore, weconsider PTB's measurements for the full temperature and owrate range to be valid.

    In the case of NMIJ, similar arguments can be presented by themeasurement of a ow nozzle as shown in the Ref. [25]. Thedischarge coefcients measured are on the same curve as a

    Fig. 20. Measurement results for the I-UFM at PTB and NMIJ. (a) Path 1, (b) path 5, (c) path 2, (d) path 4, (e) path 3, (f) sum, (g) RMS difference between PTB and NMIJ acrossow rates, and (h) path expected errors.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 2842 41

  • function of the Reynolds number not only for 20 1C and 40 1C butalso for higher temperatures. Consequently, the measurements inNMIJ for the full temperature and ow rate ranges are also valid.

    Given the fact that the degree of complexity of the geometry of adiametral path is lower than the complexity of the geometry of anorice plate and its corresponding taping systems, it is believed thatthe UFM will be capable of improving the uncertainty provided byorice plates.

    Based on the actual experiences, it should be taken intoconsideration that a UFM based ow measurement device thatcould be used to extrapolate the result to conditions outside thecalibration ranges should be based on one or more central paths,provided that predictable ow proles exist, as for example after along inlet pipe, or after a diameter reduction.

    6. Conclusions

    The central path of the UFM has fullled the conditions to serveas a transfer standard. It has a good repeatability, and provided thesame ow prole is given, also has a good reproducibility. Theerror introduced by path 3 is dependent on the Reynolds number.This is the basis for any similarity based extrapolation.

    The weighted sum used in the UFM is a robust method tocompensate for asymmetries and for errors introduced by thedifferent paths. Since non-Reynolds-dependent errors are mutuallycancelled, the result of a weighted summation appears to be, tosome extent, only Reynolds dependent. This technique is the bestchoice if nearly fully developed ow conditions cannot be reachedand a reproducibility of about 0.15% is sufcient.

    Analyzing the performance of ow meters it is of great value ifit can be guaranteed, for example via UVP or LDV, that fullydeveloped ow conditions exist.

    7. Further work

    There are several open questions about the errors introducedby the pockets. It has been conrmed that the error of the centralpath is Reynolds dependent, but its exact description has not beenperformed. Using the WCH, a measurement campaign will bestarted to characterize the behavior of the error of the central pathwith aid of UVP and LDV. The experiences on path 3 will be thebasis for a later characterization of the outer paths.

    Acknowledgments

    The generous cooperation of the KROHNE Company whichprovided, installed and congured the ultrasonic systems in thewindow chamber is greatly appreciated, as well as the activecollaboration of Konstantin Richter during the measurement cam-paigns in Berlin.

    References

    [1] Shimada T, Doihara R, Terao Y, Takamoto M. Development of primary standardfor hydrocarbon ow and traceability system of measurement in Japan.Synthesiology 65/110English edition 2010;3(1).

    [2] Lunde P, Frysa KE, Vestrheim M. GERG project on ultrasonic gas ow meters,phase II technical report. Groupe Europeen de recherches gazieres; 2000.

    [3] Hydraulic turbines and pump-turbines. PTC 18-2011. ASME, New York; 2011.[4] ISO 12242:2012. Measurement of uid ow in closed conduitsultrasonic

    transit-time meters for liquid; 2012.[5] AGA Report 9. Measurement of gas by multipath ultrasonic meters; 1998.[6] ISO 5167-2:2003. Measurement of uid ow by means of pressure differential

    devices inserted in circular cross-section conduits running fullPart 2: Oriceplates; 2003.

    [7] Moore PI, Brown GJ, Stimpson BP. Ultrasonic transit-time owmeters mod-elled with theoretical velocity proles: methodology. Meas Sci Technol2000;11:1802.

    [8] Voser Alexandre. Analyse und Fehleroptimierung der mehrpfadigen akus-tischen Durchussmesung in Wasserkraftanlagen. Dissertation ETH No. 13102,Zurich 1999.

    [9] Pannell CN, Evans WAB, Jackson DA. A new integration technique forowmeters with chordal paths. Flow Meas Instrum 1990;216(1):224.

    [10] Lau Peter, Stolt Krister. Calibration intercomparison on ow meter forkerosene synthesis report. Swedish National Testing and Research InstituteSP Report; 1995. p. 77. ISBN 91-7848-606-8.

    [11] Paton Richard. Final report on international key comparison of liquid hydro-carbon ow facilities. CCM-FF-K2 2008 Metrologia 45, 07019.

    [12] Tawackolian K, Bker O, Hogendoorn J, Lederer T. Investigation of a ten-pathultrasonic ow meter for accurate feedwater measurements. Meas Sci Technol2014;25:075304.

    [13] Zheng D, Zhang P, Xu T. Study of acoustic transducer protrusion and recesseffects on ultrasonic owmeter measurement by numerical simulation. FlowMeas Instrum 2011;22:48893.

    [14] Yeh TT, Mattingly GE. Computer simulation of ultrasonic ow meter perfor-mance in ideal and non-ideal pipeows. In: Proceedings of ASME FEDSM'97;1997.

    [15] Furuichi N, Terao Y, Takamoto M. A new calibration facility of ow rate for highReynolds number. Flow Meas Instrum 2009;20(1):3847.

    [16] Mathies N. Messunsicherheit einer gravimetrischen Kalt- und Warmwasser-Normalmessanlage fr groe Volumenstrme [Dissertation]. Technische Uni-versitt Berlin. Berlin: Mensch & Buch Verlag; 2005.

    [17] Thorns J. Analytische und experimentelle Untersuchung der Messunsicherheiteines Geschwindigkeitsnormals zur Kalibrierung von LDV-systemen [Master'sthesis]. Technical University of Berlin; 2010.

    [18] Yeh TT, Mattingly GE. Pipeow downstream of a reducer and its effects onow meters. Flow Meas Instrum 1994;5:S1817.

    [19] Gersten K. Fully developed pipe ow. In: Merzkirch W, editor. Fluid mechanicsof ow metering. Berlin, Heidelberg: Springer-Verlag; 2005.

    [20] Richtlinie zur strmungstechnischen Validierung von Kalibrier-Prfstndenim Rahmen der EN 1434; October 2009.

    [21] Drenthen J, Kurzt M, van Klooster J, Vermeulen M. Reducing installationeffects on ultrasonic ow meters. In: The seventh international symposium onuid ow measurement; 2009 .

    [22] Cousins T, Estrada H, Augenstein D. Installation effects and diagnostic inter-pretation using the Caldon ultrasonic meter. In: North sea ow measurementworkshop, St Andrews, Scotland; October 2004.

    [23] Mattingly GE, Yeh TT. Flow meter Installation Effects due to several elbowcongurations. In: Proceeding of the second international symposium on uidow measurement, Galgary, Alberta, Canada; 1990. p. 27183.

    [24] Brown GJ, Grifth BW. A New ow conditioner for 4-path ultrasonic owmeters. In: Proceedings of the FLOMEKO 2013, Paris; 2013.

    [25] Furuichi N, Cheong KH, Terao Y, Nakao S, Fujita K, Shibuya K. Experimentalresults of ow nozzle based on PTC 6 for high Reynolds number. In:Proceedings of the ASME 2014 power conference, POWER2014-32116, July2831, 2014, Baltimore, Maryland, USA.

    [26] Takeda Y. Ultrasonic doppler velocity proler for uid ow. 1st ed.. Tokyo:Springer-Verlag; 2012.

    L. Cordova et al. / Flow Measurement and Instrumentation 45 (2015) 284242

    Qualification of an ultrasonic flow meter as a transfer standard for measurements at Reynolds numbers up to 4times106...Introduction and motivationTraceability of flow meters outside calibration rangesUltrasonic flow metersIdeal case integrationReal case traceability limits

    The flow test rigsFlow rate facility NMIJFlow rate facility PTBInternal consistency test for the flow calibration facility

    MethodsPreliminary measurementsVelocity profile measurementThe window chamberThe flow profileMeasurements of the TB profileRadial components on the central path

    Single path measurementsDesign of the comparison

    ResultsMeasurement conditionsPath projection resultsPath by path comparison results

    DiscussionConclusionsFurther workAcknowledgmentsReferences